Course Introduction
Regression Kink Design for Public Policy Evaluation Training Course equips economists, policy researchers, impact evaluators, and quantitative analysts with advanced econometric methods for estimating causal effects from policy rules that create changes in the slope of an assignment or treatment function. Public policies frequently determine benefits, tax liabilities, subsidies, insurance coverage, pension contributions, or programme entitlements through formulas that vary with income, age, earnings, household composition, or other measurable characteristics. When these rules contain a known kink, the relationship between an assignment variable and treatment intensity changes at a specific threshold, creating an opportunity to investigate causal effects under appropriate identifying assumptions. This course provides a rigorous framework for recognizing such opportunities, formulating regression kink research designs, estimating treatment effects, validating assumptions, and translating empirical findings into credible policy evidence.
Many public programmes allocate benefits through continuous formulas rather than simple eligibility cutoffs. For example, a benefit may increase with earnings at one rate below a threshold and at a different rate above it, or a subsidy may gradually decline as household income rises. These rules can generate changes in the slope of treatment exposure without necessarily producing a discontinuous jump in treatment levels. Regression Kink Design (RKD) uses this change in slope to identify causal effects when the underlying relationship between the outcome and the assignment variable would otherwise evolve smoothly through the kink. Participants learn how to distinguish regression kink designs from regression discontinuity designs, identify suitable policy formulas, determine whether the observed kink creates useful identifying variation, and evaluate whether the policy context supports a credible counterfactual comparison.
The course examines the theoretical foundations of regression kink estimation, including potential outcomes, treatment assignment mechanisms, continuity and smoothness assumptions, treatment intensity, and the relationship between the assignment rule and the outcome response. Participants explore the sharp regression kink design, where treatment exposure follows a known kinked assignment function, and fuzzy regression kink designs, where the treatment response exhibits a change in slope rather than perfectly following the policy rule. They learn how the reduced-form change in the outcome slope relates to the change in the treatment slope, how local causal effects are interpreted, and why identification can fail when other determinants of the outcome change slope at the same threshold. Particular attention is given to the importance of differentiability assumptions, the scale and precision of the running variable, and the credibility of the policy mechanism underlying the design.
Reliable regression kink analysis requires careful data preparation, bandwidth selection, functional-form decisions, and inference. A visually apparent kink does not automatically establish a causal relationship, and conventional regression specifications may be sensitive to polynomial order, bandwidth choice, measurement error, or limited observations near the threshold. Participants learn to estimate local linear and higher-order specifications where appropriate, compare alternative bandwidths, examine sensitivity to covariates, and investigate whether the assignment variable has sufficient support around the kink. They also explore robust bias-corrected inference, heteroskedasticity-consistent standard errors, clustering, placebo thresholds, density and sorting diagnostics, and falsification tests. The programme emphasizes that diagnostic results must be interpreted in conjunction with substantive knowledge of the policy assignment mechanism and cannot independently prove the validity of the identifying assumptions.
The programme applies regression kink methods to practical policy questions involving income-tested benefits, tax schedules, unemployment insurance, pension formulas, social protection transfers, education funding, agricultural subsidies, healthcare financing, and other programmes that vary continuously with measurable characteristics. Participants learn to define the relevant policy parameter, select outcome measures, construct the running variable and treatment intensity, examine administrative rules, and identify potential confounding mechanisms. They also investigate extensions involving multiple kinks, fuzzy assignment mechanisms, heterogeneous treatment responses, panel data, and combinations with other quasi-experimental approaches. Through applied exercises, participants develop the ability to distinguish a theoretically eligible policy formula from an empirically informative design and to determine when alternative causal methods may be more defensible.
By completing this course, participants will be able to identify, implement, test, and communicate regression kink evaluations with greater methodological rigor. They will develop practical skills in policy-rule analysis, causal identification, local estimation, bandwidth selection, robust inference, sensitivity analysis, and transparent reporting. Through a capstone project, participants will construct an RKD research design, estimate local causal effects, evaluate identifying assumptions, and prepare a technical report that explains the results and their limitations. The programme supports public institutions, development organizations, research centres, and policy evaluation teams seeking to generate credible evidence about how changes in policy incentives and benefit formulas influence individual behaviour, programme participation, and public policy outcomes.
Duration
10 days
Who Should Attend
-
Monitoring, evaluation, research, and learning professionals conducting advanced quantitative policy and programme evaluations.
-
Economists and econometricians studying taxation, social protection, labour markets, public finance, and public-sector programme design.
-
Impact evaluation specialists developing quasi-experimental research strategies for policy interventions with formula-based assignment mechanisms.
-
Government policy analysts evaluating tax schedules, benefit formulas, subsidy programmes, pension systems, and income-related entitlements.
-
Development researchers working with multilateral development banks, bilateral donors, international organizations, and research institutions.
-
Quantitative social scientists investigating behavioural responses to financial incentives, eligibility formulas, and changes in treatment intensity.
-
Public finance specialists assessing the effects of marginal tax rates, transfer schedules, and fiscal policy incentives.
-
Social protection analysts evaluating unemployment insurance, cash transfers, disability benefits, and income-tested programmes.
-
Health and education researchers examining funding formulas, insurance contributions, service subsidies, and other continuous policy allocation mechanisms.
-
Data scientists and statistical analysts working with administrative records, longitudinal datasets, and econometric evaluation models.
-
Academic researchers and postgraduate students seeking advanced training in causal inference and applied econometric methods.
-
Evaluation consultants preparing technical studies, policy briefs, impact assessments, and research reports.
-
Statistical officers responsible for preparing and validating administrative datasets used in policy impact evaluation.
-
Evaluation directors and technical advisers reviewing identification strategies, econometric specifications, and causal evidence.
-
Professionals seeking advanced expertise in regression kink design, regression discontinuity, and credible quasi-experimental evaluation.
Course Objectives
-
Develop an advanced understanding of regression kink design and its application to policies where treatment assignment or intensity changes slope at a known threshold.
-
Distinguish regression kink design from regression discontinuity, instrumental variable estimation, interrupted time series, and conventional regression analysis.
-
Identify suitable policy rules and assignment mechanisms by examining benefit formulas, tax schedules, subsidies, contribution rates, and other continuous allocation systems.
-
Formulate causal identification strategies using potential outcomes, policy assignment functions, smoothness assumptions, and clearly defined local treatment effects.
-
Construct appropriate running variables, treatment measures, outcome variables, bandwidths, and analytical samples for sharp and fuzzy regression kink designs.
-
Implement local linear and other defensible regression specifications while understanding the implications of polynomial order, functional form, and limited observations near the kink.
-
Apply bandwidth selection, robust bias-corrected inference, appropriate standard errors, and sensitivity analysis to improve the reliability of estimated treatment effects.
-
Assess identifying assumptions by examining alternative sources of slope changes, sorting, measurement error, concurrent policies, and changes in the composition of observations near the threshold.
-
Conduct placebo tests, falsification analyses, alternative bandwidth checks, and specification comparisons to evaluate the robustness of regression kink findings.
-
Interpret sharp and fuzzy regression kink estimates, including their local nature, treatment intensity implications, uncertainty, and limitations in generalizing to wider populations.
-
Use statistical software and reproducible analytical workflows to estimate, validate, visualize, and document regression kink models for applied policy research.
-
Communicate findings responsibly by connecting empirical estimates to policy mechanisms, identifying assumptions, practical significance, uncertainty, and implications for policy design.
Comprehensive Course Outline
Module 1: Foundations of Regression Kink Design and Causal Policy Evaluation
-
Examine causal inference principles, potential outcomes, counterfactual reasoning, treatment assignment, and the distinction between observed associations and causal effects.
-
Explain how changes in the slope of policy assignment functions can generate identifying variation for estimating the effects of treatment intensity.
-
Compare regression kink design with regression discontinuity, instrumental variables, difference-in-differences, and other quasi-experimental approaches.
-
Identify policy settings suitable for RKD and distinguish a genuine assignment-rule kink from an apparent slope change caused by data limitations or unrelated factors.
Module 2: Policy Assignment Rules, Running Variables, and Kink Identification
-
Identify policy formulas involving income, earnings, age, contribution levels, benefit schedules, subsidy rates, or other continuous assignment variables.
-
Define the running variable, kink threshold, treatment intensity, outcomes, observation periods, and relevant policy parameters for the evaluation.
-
Examine administrative rules and institutional documentation to determine whether the assignment function contains a known and substantively meaningful change in slope.
-
Assess the availability, precision, range, and measurement quality of the running variable and determine whether sufficient observations exist around the kink.
Module 3: Theoretical Foundations and Identification Assumptions
-
Develop the potential outcomes framework for regression kink analysis and explain the conditions under which changes in outcome slopes can identify causal effects.
-
Examine continuity and smoothness assumptions governing the untreated outcome relationship and other relevant characteristics around the policy kink.
-
Distinguish changes in the treatment assignment slope from changes in the outcome slope and explain how their relationship supports causal identification.
-
Identify violations involving concurrent policy rules, discontinuous institutional changes, sorting, compositional shifts, and other factors that can generate misleading kink estimates.
Module 4: Sharp Regression Kink Design
-
Examine sharp RKD settings in which treatment exposure follows a known assignment function with a change in slope at the specified threshold.
-
Derive and interpret the sharp regression kink estimand using the relationship between the outcome slope change and the treatment assignment slope change.
-
Implement local regression models around the kink and assess how the estimated effect depends on the selected neighborhood and functional specification.
-
Interpret sharp RKD results in terms of local causal effects, treatment intensity, policy incentives, and the population represented by observations near the kink.
Module 5: Fuzzy Regression Kink Design and Instrumental Variable Connections
-
Examine fuzzy RKD settings where actual treatment exposure responds imperfectly to a policy rule and the treatment relationship changes slope at the kink.
-
Explain how the policy assignment mechanism can provide an instrument-like source of variation for treatment intensity under appropriate assumptions.
-
Estimate the relationship between outcome slope changes and treatment slope changes while accounting for imperfect compliance or incomplete treatment responses.
-
Assess the interpretation and limitations of fuzzy RKD estimates, including treatment heterogeneity, alternative causal pathways, and the assumptions required for identification.
Module 6: Local Regression Estimation and Functional Form Selection
-
Implement local linear and defensible higher-order regression specifications for estimating outcome and treatment slopes around the policy threshold.
-
Compare kernel weighting, local polynomial methods, covariate adjustment, and alternative functional forms while recognizing the limits of global polynomial specifications.
-
Examine how polynomial order, bandwidth choice, predictor inclusion, and data support influence estimated slope changes and uncertainty.
-
Establish a transparent specification strategy that prioritizes credible local estimation, interpretable assumptions, and robustness over arbitrary model complexity.
Module 7: Bandwidth Selection, Data Support, and Robust Inference
-
Understand the role of bandwidth in balancing local comparability, statistical precision, functional-form bias, and the amount of usable data around the kink.
-
Apply suitable bandwidth selection procedures and compare results across alternative windows to assess the stability of the estimated effect.
-
Implement heteroskedasticity-consistent standard errors, robust bias-corrected inference, and clustering procedures appropriate to the data structure and assignment mechanism.
-
Evaluate the implications of sparse observations, heaping, rounding, measurement error, and limited variation in the running variable for statistical reliability.
Module 8: Validity Diagnostics and Identification Threats
-
Investigate whether other policy provisions, institutional processes, market conditions, or administrative practices create simultaneous changes in outcome determinants at the kink.
-
Examine the distribution of the running variable, potential manipulation, bunching, sorting, and changes in observation composition around the threshold.
-
Assess continuity in predetermined covariates and apply appropriate placebo thresholds, alternative outcomes, and falsification procedures.
-
Interpret diagnostic evidence cautiously, recognizing that successful balance or placebo tests cannot prove all identifying assumptions or exclude every alternative explanation.
Module 9: Sensitivity Analysis and Specification Robustness
-
Compare estimates across alternative bandwidths, kernel choices, polynomial orders, covariate specifications, outcome definitions, and sample restrictions.
-
Investigate whether estimated effects are driven by influential observations, data heaping, measurement anomalies, structural changes, or narrow segments of the running variable.
-
Develop sensitivity frameworks that document how estimates and confidence intervals respond to reasonable analytical decisions.
-
Establish transparent reporting standards for distinguishing stable findings from specification-sensitive results and unresolved identification concerns.
Module 10: Treatment Heterogeneity, Interpretation, and External Validity
-
Examine how treatment responses may differ across individuals, households, firms, institutions, geographic settings, and baseline characteristics near the kink.
-
Distinguish local effects identified by the design from average treatment effects for the full programme population or wider policy environment.
-
Investigate how changes in treatment intensity, compliance, eligibility, or behavioural response influence the practical interpretation of RKD estimates.
-
Assess whether local findings are informative for policy reforms affecting populations or assignment ranges substantially different from those represented near the kink.
Module 11: Panel Data, Repeated Observations, and Multiple Kinks
-
Examine the use of repeated observations and panel data for studying outcomes before and after policy-rule changes while accounting for the assumptions required by the chosen design.
-
Analyze policies containing multiple kinks and determine whether each threshold represents a distinct treatment mechanism or interacts with other provisions.
-
Assess changes in policy formulas over time and evaluate whether multiple thresholds, reforms, or changing administrative practices complicate identification.
-
Apply appropriate inference and sensitivity procedures for clustered observations, repeated measurements, multiple thresholds, and correlated policy exposure.
Module 12: Regression Kink Design in Public Finance and Social Protection
-
Evaluate tax schedules, unemployment insurance, pension contributions, income-tested benefits, and social assistance formulas that generate slope changes in financial incentives.
-
Analyze how marginal tax rates, benefit withdrawal rates, subsidy schedules, and contribution formulas may influence employment, earnings, participation, and household behaviour.
-
Identify relevant administrative records and outcome measures while assessing privacy, data quality, measurement consistency, and policy implementation details.
-
Translate local treatment effect estimates into policy-relevant interpretations while distinguishing behavioural responses from broader welfare or fiscal conclusions requiring additional analysis.
Module 13: Applications in Health, Education, Agriculture, and Public Services
-
Examine RKD applications involving healthcare contributions, insurance benefits, education funding formulas, agricultural subsidies, and other continuous public resource allocation rules.
-
Select outcome indicators and analytical periods that correspond to the mechanisms through which changes in policy incentives may influence behaviour or service delivery.
-
Assess whether institutional differences, concurrent reforms, geographical factors, or implementation variation undermine the comparability needed for credible identification.
-
Develop applied evaluation strategies that connect econometric estimates with programme objectives, service outcomes, distributional implications, and implementation realities.
Module 14: Statistical Software and Reproducible RKD Workflows
-
Implement regression kink estimation and associated diagnostics using statistical software such as R or Stata and suitable econometric packages.
-
Develop reproducible scripts for constructing running variables, identifying thresholds, preparing analytical samples, selecting bandwidths, estimating models, and producing diagnostic outputs.
-
Create visualizations of policy assignment functions, treatment relationships, outcome trends, local regressions, and estimated slope changes around the kink.
-
Establish quality assurance procedures for data validation, code review, software version control, model documentation, independent replication, and transparent reporting.
Module 15: Advanced Extensions, Emerging Methods, and Policy Communication
-
Examine methodological developments involving robust inference, alternative bandwidth procedures, multiple kinks, fuzzy treatment mechanisms, and extensions to more complex policy assignment settings.
-
Assess the compatibility and limitations of combining RKD with panel methods, instrumental variable strategies, other quasi-experimental designs, and machine-assisted analytical workflows.
-
Develop technical reporting standards that document the assignment rule, identifying assumptions, estimator choices, diagnostics, sensitivity checks, and uncertainty.
-
Communicate results to policymakers using clear effect interpretations, visual evidence, practical significance, methodological caveats, and appropriately qualified policy recommendations.
Module 16: Integrated Regression Kink Evaluation Capstone
-
Develop a complete RKD evaluation design for a realistic policy formula, specifying the running variable, kink point, treatment measure, outcome, estimand, and identifying assumptions.
-
Prepare and analyze an applied dataset, estimate the relevant outcome and treatment slope changes, and implement appropriate local regression procedures.
-
Conduct bandwidth sensitivity analysis, validity diagnostics, placebo tests, specification comparisons, and robust inference to assess the credibility of the results.
-
Present a technical report and policy briefing explaining the identification strategy, estimated effects, robustness evidence, unresolved limitations, and implications for policy evaluation.
Training Approach
This course will be delivered by our skilled trainers who have vast knowledge and experience as expert professionals in the fields. The course is taught in English and through a mix of theory, practical activities, group discussion and case studies. Course manuals and additional training materials will be provided to the participants upon completion of the training.
Tailor-Made Course
This course can also be tailor-made to meet organization requirement. For further inquiries, please contact us on: Email: training@upskilldevelopment.com Tel: +254 721 331 808
Training Venue
The training will be held at our Upskill Training Centre. We also offer training for a group (at a discount of 10% to 50%) at requested location all over the world. The Onsite course fee covers the course tuition, training materials, two break refreshments, buffet lunch, airport transfers, Upskill gift package, and guided tour.
Visa application, travel expenses, dinners, accommodation, insurance, and other personal expenses are catered by the participant
Certification
Participants will be issued with Upskill certificate upon completion of this course.
Airport Pickup and Accommodation
Airport pickup and accommodation is arranged upon request. For booking contact our Training Coordinator through Email: training@upskilldevelopment.com, +254 721 331 808
Terms of Payment:
Unless otherwise agreed between the two parties’ payment of the course fee should be done 3 working days before commencement of the training so as to enable us to prepare better.








